Integration of graph neural networks and long short-term memory models for advancing heart failure prediction

Abstract

Heart failure constitutes a chronic disease affecting millions of people worldwide, hence creating an important burden on healthcare infrastructures. Predictive models about the onset or worsening of HF can be instrumental in conducting proper and timely interventions to improve the outcomes of the care of patients with HF. This paper introduces a novel approach to predicting HF, integrating graph neural networks (GNNs) with long short-term memory (LSTM) networks for better prediction accuracy. This hybrid model, GNN-LSTM, applies the advantages of both networks: the complex interdependencies between clinical variables capture clinical relationships; LSTMs can better manage temporal dependencies. The model was tested on a large, highly representative dataset containing diversified clinical variables from HF patients, with 98.9 % predictive accuracy, which outperforms the single models as well as their respective performances by conventional methods like CNN, SMOTE, LSTM-RNN, CNN-LSTM, CNN-GRU, and traditional GNN approaches. Thus, the GNN-LSTM model, developed in Python, produces robust results across cases, irrespective of coronary heart disease co-presence comorbidity. Nonetheless, one of the limitations of the research is that generability is still in the future. This integrated approach has huge promises for improving HF prediction, with early interventions and personalized health strategies that would diminish the burden on patients and healthcare systems.

Keywords

• Heart failure prediction
• Graph neural networks
• Long short-term memory
• Hybrid GNN-LSTM model
• Temporal dependencies
• Predictive accuracy